Book Review: Artificial Intelligence: A Guide for Thinking Humans

This week's review is on Artificial Intelligence: A Guide for Thinking Humans by Melanie Mitchell.

A book written by an expert on AI - Artificial Intelligence: A Guide for Thinking Humans - feels like an analysis of European diplomatic history published in 1912 by a German foreign minister. The expertise is evident, the history is clear, the explanations and examples abound and enrich, but the story feels incomplete from the other end of history. The text isn't out of date, so much given its publication in 2019, that it cannot tell the story of AI fully to those of us in the post-GPT world.

A great cover.

Mitchell ably interweaves philosophy, examples, history, and computer programming into a fully-developed conversation of what AI is, the divide between symbolic and deep learning, the primary difficulties that AI has in interacting with the "real" world, and the pace of research. Her expertise adds to the book, rather than distracts - enabling us to have a front-eye view of the AI growth from the 1980s to the 20-teens. Neatly divided into five sections and following a roughly chronological outline, Mitchell's book is one that ultimately doesn't make a bold claim but points at the uniqueness of human intelligence apart from machines.

Beginning with a presentation at Google in 2014* (an invitee of renowned scientist and writer, Douglas Hofstadter), Mitchell traces her journey into the progress of AI in the 20-teens while calling back to its post-war origins. She does so in a way that evokes both the milestone efforts (ELIZA, Deep Blue, Watson, et al.) and the little steps and teams along the way that grew both the symbolic and deep learning approaches. She does so in an effort "for you to share in my own exploration and, like me, to come away with a clearer sense of what the field has accomplished and how much further there is to go before our machines can argue for their own humanity" (14).

Although at times, her tone is professorial ("I'll explain shortly" pops up repeatedly), the somewhat conversational tone of the book allows complexity in thought while permitting simplicity in language - making it more accessible. In charting the shifting boundaries of what AI can and cannot do - "Easy things are hard" (33) - Mitchell enables a philosophical consideration of what consciousness, creativity, and, ultimately, humanity means.

She does an excellent job of charting how we shift what seems important in these realms - chess is key until Deep Blue defeats Kasparov; Go matters until AlphaGo triumphs, et al. In doing so she highlights John McCarthy's maxim: "As soon as it works, no one calls it AI anymore?" This tendency is something I notice in myself - LLMs certainly have their limitations but the ability to generate text and art of moderate capability for minimal cost is something sci-fi only a decade ago.

In the end, the necessity of humanity remaining "in the loop" is a core message of the book. From the final chapter - a series of shorter responses to deeper questions - drawing from Sendhil Mullainathan:

We should be afraid. Not of intelligent machines. But of machines making decisions that they do not have the intelligence to make. I am far more afraid of machine stupidity than machine intelligence...

This analysis and Mitchell's adherence to a more-total symbolic approach while remaining skeptical to deeper learning and the GPT-era places her closer to Gary Marcus and highlights her occasionally published Substack posts. This skepticism is not as present in the text, but serves as one of the currents which Mitchell brings to her deeper discussion of the uniqueness of human intelligence in an era of growing artificial intelligence.

Artificial Intelligence: A Guide for Thinking Humans by Melanie Mitchell.

Rating: 4/5 Stars

Good For: An introduction to AI, the philosophical problems, and landscape just prior to the Large Language Model boom.

Best nugget: "Easy things are hard" for AI. What might be easy for a human - scanning a picture for particular elements or driving a car; requires massive processing and data capabilities. What might be easy for a machine - multiplying multi-digit numbers - does not.

Please note: As an Amazon Associate I earn from qualifying purchases. However, I am not paid to provide reviews or use content.

*"In short, Google is no longer merely a web-search portal...It is rapidly becoming an applied AI company. AI is the glue that unifies the diverse products, services, and blue-sky research efforts offered by Google." (4)